An intelligent identification method and system for pine wood nematode disease based on drone technology

Abstract

The present invention discloses a method and system for intelligent identification of pine wilt disease based on unmanned aerial vehicle technology, including: S1: using an unmanned aerial vehicle equipped with a multispectral camera to collect images of tree areas, applying a bilateral filtering algorithm to remove image noise, and obtaining a denoised image; S2: extracting color features, texture features, and vegetation index features from the denoised image; S3: manually annotating the denoised image to mark the tree area infected with pine wilt disease; S4: integrating the denoised image, color features, texture features, vegetation index features, and segmentation masks, and constructing and training a pine wilt disease segmentation model based on U‑Net; S5: applying the trained pine wilt disease segmentation model to identify the pine wilt disease infected area. The present invention can provide an efficient and accurate method for intelligent identification of pine wilt disease, which utilizes modern unmanned aerial vehicle technology and deep learning algorithms to overcome the limitations of traditional monitoring methods.

Description

Technical Field

[0001] The present invention relates to the technical field of pine wood nematode disease identification, and in particular to an intelligent pine wood nematode disease identification method and system based on unmanned aerial vehicle technology. Background Art

[0002] Pine wilt disease is a serious forestry disease that seriously affects the health and economic value of forest ecosystems. Traditional pine wilt disease monitoring methods mainly rely on manual inspections and laboratory testing, which is not only time-consuming and labor-intensive, but also difficult to achieve large-scale real-time monitoring. With the development of drone technology and image processing technology, using drones equipped with multispectral cameras to collect images of tree areas and identifying pine wilt disease infected areas through image processing technology has become an effective solution.

[0003] The invention patent with application number 202110321019.6 discloses a method and device for image recognition and detection of pine wood nematode disease. The method uses a drone equipped with a digital camera to obtain images of the experimental area, performs image enhancement through the AS-SMOTE Boost algorithm, and uses the bispectral grayscale image to extract feature weights for denoising. Finally, the pine wood nematode disease is identified by extracting and classifying morphological features, color features, and texture features. Although this method has achieved basic automatic recognition functions, it still has the following problems: (1) Only ordinary digital cameras are used to obtain images, lacking multispectral information, resulting in incomplete feature extraction; (2) Important feature information such as vegetation index is not fully utilized during feature extraction; (3) The recognition model structure is relatively simple and fails to fully utilize the advantages of deep learning, resulting in limited recognition accuracy. Summary of the invention

[0004] In view of this, the present invention provides a method and system for intelligent identification of pine wood nematode disease based on drone technology, aiming to provide an efficient and accurate method for intelligent identification of pine wood nematode disease, using modern drone technology and deep learning algorithms to overcome the limitations of traditional monitoring methods.

[0005] To achieve the above purpose, the present invention provides an intelligent identification method for pine wood nematode disease based on drone technology, comprising the following steps:

[0006] S1: Use a drone equipped with a multispectral camera to collect images of the tree area, apply a bilateral filtering algorithm to denoise the image, and obtain a denoised image;

[0007] S2: Extract color features, texture features and vegetation index features from the denoised image; the texture features include contrast and energy ; The vegetation index feature is the normalized difference vegetation index ;

[0008] S3: Manually annotate the denoised image, mark the areas infected with pine wilt disease, and generate segmentation masks ,in , Indicates pixel position It is an area infected with pine wood nematode disease. Indicates pixel position It is not an area infected by pine wood nematode disease;

[0009] S4: Integrate the denoised image, color features, texture features, vegetation index features and segmentation masks, and build and train a pine wood nematode segmentation model based on U-Net;

[0010] The step S4 includes the following steps:

[0011] S41: Integrate the denoised image, color features, texture features, vegetation index features and the segmentation mask in step S3 into a training set. Each sample in the training set is specifically:

[0012] ;

[0013] in, is the input image, including the denoised image , Color Features , Contrast ,energy and Normalized Difference Vegetation Index ;

[0014] S42: Building a pine wilt disease segmentation model based on the U-Net architecture. The pine wilt disease segmentation model consists of an encoder and a decoder, and a jump connection is added between the encoder and the decoder.

[0015] S43: Use the training set data to train the pine wood nematode segmentation model and use the stochastic gradient descent algorithm to minimize the loss function , obtain the trained pine wood nematode disease segmentation model;

[0016] S5: Use the multispectral camera carried by the drone to collect new tree area images, denoise the new tree area images and extract color features, texture features and vegetation index features; apply the trained pine wilt disease segmentation model to identify the pine wilt disease infected area.

[0017] As a further improvement method of the present invention:

[0018] Optionally, in step S1, using a multispectral camera mounted on an unmanned aerial vehicle to collect an image of a tree area, applying a bilateral filtering algorithm to denoise the image, and obtaining a denoised image includes the following steps:

[0019] S11: Using a multispectral camera mounted on a drone to capture images of tree areas ;

[0020] S12: Apply bilateral filtering algorithm to the image of tree area Perform denoising to obtain the denoised image , specifically:

[0021] ;

[0022] in, and Images of tree areas The horizontal pixel position and the vertical pixel position of The denoised image At pixel position The pixel value at ; Image of the tree area At pixel position The pixel value at ; and are the horizontal and vertical offsets respectively, and Traverse within the filter window; is the size of the filter window; Image of the tree area At pixel position The pixel value at ; is the spatial Gaussian kernel function; is the range Gaussian kernel function.

[0023] Optionally, extracting color features, texture features and vegetation index features from the denoised image in step S2 includes the following steps:

[0024] S21: Use color space transformation to transform the denoised image The red, green, and blue bands in the image are transformed from the RGB color space to the HSV color space to obtain color features. ;

[0025] S22: Extract texture features:

[0026] S221: For each pixel position , construct the local gray-level co-occurrence matrix , specifically:

[0027] ;

[0028] in, is the element index in the local gray-level co-occurrence matrix, and Corresponding to the two gray levels in the denoised image, ranging from 0 to An integer between is the number of gray levels; At pixel position of In the neighborhood, the gray level is and The frequency of pixel pairs; For Centered A collection of pixel position indices within the range. for The pixel position index in ; The denoised image At pixel position The pixel value at ; is the indicator function, when 1 when it is, otherwise 0;

[0029] S222: Extracting texture features based on the local gray level co-occurrence matrix, wherein the texture features include contrast and energy , specifically:

[0030] ;

[0031] ;

[0032] in, For contrast At pixel position The contrast value at For energy At pixel position The energy value at

[0033] S23: Extracting vegetation index features, where the vegetation index features are normalized difference vegetation index , specifically:

[0034] ;

[0035] in, Normalized Difference Vegetation Index At pixel position The value at and They are the denoised images respectively. Near infrared and red light bands; and They are and At pixel position The pixel value at .

[0036] Optionally, in step S4, the loss function is specifically:

[0037] ;

[0038] in, is the segmentation mask predicted by the pine wilt segmentation model At pixel position The value at ; and They are the denoised images respectively. The number of pixels in the horizontal direction and the number of pixels in the vertical direction.

[0039] Optionally, in step S5, a multispectral camera carried by a drone is used to collect new tree area images, and the new tree area images are denoised and color features, texture features, and vegetation index features are extracted; and the trained pine wilt disease segmentation model is applied to identify the pine wilt disease infected area, including the following steps:

[0040] Using a multispectral camera carried by a drone to collect a new tree area image, performing denoising on the new tree area image in step S1 to obtain a denoised image, and then performing color feature, texture feature and vegetation index feature extraction on the denoised image in step S2; integrating the denoised image obtained using the new tree area image with the color feature, texture feature and vegetation index feature to obtain a new input image;

[0041] The trained pine wilt disease segmentation model is used to segment the new input image. The pine wilt disease segmentation model outputs a predicted segmentation mask. The pixel position with a pixel value of 1 in the segmentation mask is the identified pine wilt disease infected area.

[0042] The present invention also discloses an intelligent identification system for pine wood nematode disease based on drone technology, comprising:

[0043] Collection and denoising module: Use a drone equipped with a multispectral camera to collect images of the tree area, and apply a bilateral filtering algorithm to denoise the images;

[0044] Feature extraction module: extracts color features, texture features and vegetation index features from the denoised image;

[0045] Annotation module: manually annotate the denoised image and mark the areas infected with pine wood nematode disease;

[0046] Model building module: Integrate denoised images, color features, texture features, vegetation index features and segmentation masks, build and train a pine wood nematode segmentation model based on U-Net;

[0047] Model application module: Use the trained pine wilt disease segmentation model to identify pine wilt disease infected areas.

[0048] Beneficial effects:

[0049] The present invention uses a multispectral camera equipped with an unmanned aerial vehicle to perform large-scale image acquisition, which significantly improves the efficiency and coverage of pine wood nematode disease monitoring. Traditional monitoring methods mainly rely on manual field inspections, which are not only time-consuming and labor-intensive, but also restricted by geographical conditions, making it difficult to fully cover a wide range of forest areas. The application of unmanned aerial vehicle technology allows a large area to be covered in one flight, greatly shortening the monitoring time. The use of multispectral cameras can also capture information in different bands, providing a richer data basis for subsequent image processing. This highly efficient, large-scale monitoring method can promptly detect early signs of pine wood nematode disease, gaining valuable time for timely prevention and control measures, thereby effectively curbing the spread of the disease.

[0050] The present invention utilizes a bilateral filtering algorithm to denoise the collected images and further extracts a variety of features, thereby significantly improving the accuracy and reliability of pine wilt disease identification. The bilateral filtering algorithm removes noise while retaining the image edges, improves the image quality, and makes subsequent feature extraction more accurate. In addition, color features and texture features can reflect changes in the appearance of trees, while vegetation index features can reveal the physiological state of trees. The combination of the three can capture the characteristics of the disease from different angles, further enhancing the accuracy of identification. The present invention integrates a variety of feature images, generates segmentation masks through artificial labeling, and uses the U-Net architecture to build and train a segmentation model, thereby fully utilizing the color, texture, and vegetation index features of multispectral images, enhancing the learning ability and segmentation accuracy of the model, and by minimizing the loss function, the model can more effectively identify pine wilt disease infected areas, thereby improving the reliability and robustness of overall identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a schematic diagram of a process of an intelligent identification method for pine wood nematode disease based on drone technology according to an embodiment of the present invention;

[0052] Figure 2Schematic diagram of pine wilt disease infected area segmentation according to an embodiment of the present invention, wherein Figure (a) is a new tree area image captured by a multispectral camera carried by a drone, and Figure (b) is a pine wilt disease infected area segmentation result. DETAILED DESCRIPTION

[0053] The present invention is further described below in conjunction with the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention belong to the protection scope of the present invention.

[0054] Example 1: An intelligent identification method for pine wood nematode disease based on drone technology, such as Figure 1 As shown, the following steps are included:

[0055] S1: Use a drone equipped with a multispectral camera to collect images of the tree area, apply a bilateral filtering algorithm to denoise the image, and obtain the denoised image:

[0056] S11: Using a multispectral camera mounted on a drone to capture images of tree areas ;

[0057] S12: Apply bilateral filtering algorithm to the image of tree area Perform denoising to obtain the denoised image , specifically:

[0058] ;

[0059] in, and Images of tree areas The horizontal pixel position and the vertical pixel position of The denoised image At pixel position The pixel value at ; Image of the tree area At pixel position The pixel value at ; and are the horizontal and vertical offsets respectively, and Traverse within the filter window; is the size of the filtering window, which is 5 in this embodiment; Image of the tree area At pixel position The pixel value at ; is the spatial Gaussian kernel function, which in this embodiment is , is the circumference of a circle, is a natural constant, is the standard deviation of the spatial Gaussian kernel, which is 0.67 in this embodiment; is the range Gaussian kernel function, in this embodiment , , is the standard deviation of the range Gaussian kernel, which is 10 in this embodiment.

[0060] This step uses a multispectral camera mounted on a drone and applies a bilateral filtering algorithm to effectively retain the edge and detail information of the tree area while removing image noise. This not only improves image quality, but also provides a reliable basis for subsequent feature extraction and segmentation, significantly enhancing the accuracy and reliability of identifying pine wood nematode disease. In addition, the multispectral camera can also obtain high-quality images under different lighting conditions and adapt to complex environments, further improving the stability and effectiveness of monitoring.

[0061] S2: Extract color features, texture features, and vegetation index features from the denoised image:

[0062] S21: Use color space transformation to transform the red, green, and blue bands in the denoised image from RGB color space to HSV color space to obtain color features ;

[0063] S22: Extract texture features:

[0064] S221: For each pixel position , construct the local gray-level co-occurrence matrix , specifically:

[0065] ;

[0066] in, is the element index in the local gray-level co-occurrence matrix, and Corresponding to the two gray levels in the denoised image, ranging from 0 to An integer between is the number of gray levels; At pixel position of In the neighborhood, the gray level is and The frequency of pixel pairs; For Centered A collection of pixel position indices within the range. for The pixel position index in ; The denoised image At pixel position The pixel value at ; is the indicator function, when 1 when it is, otherwise 0;

[0067] S222: Extracting texture features based on the local gray level co-occurrence matrix, wherein the texture features include contrast and energy , specifically:

[0068] ;

[0069] ;

[0070] in, For contrast At pixel position The contrast value of For energy At pixel position Energy value;

[0071] S23: Extracting vegetation index features, where the vegetation index features are normalized difference vegetation index , specifically:

[0072] ;

[0073] in, Normalized Difference Vegetation Index At pixel position The value at and They are the denoised images respectively. Near infrared and red light bands; and They are and At pixel position The pixel value at .

[0074] This step converts the denoised image into the HSV color space, extracts color features, and uses the local gray-level co-occurrence matrix to calculate texture features such as contrast and energy. It also calculates the normalized difference vegetation index as a vegetation feature. This method can fully capture the visual and physiological changes of trees, significantly improves the diversity and characterization of features, and thus enhances the accuracy and reliability of the model's recognition of pine wood nematode disease.

[0075] S3: Manually annotate the denoised image, mark the area infected with pine wood nematode disease, and generate a segmentation mask;

[0076] Trained professionals check each denoised image one by one , for the denoised image Perform manual annotation to mark the denoised image Generate segmentation masks for areas infected with pine wilt disease ,in Indicates pixel position It is an area infected with pine wood nematode disease. Indicates pixel position It is not an area infected by pine wood nematode disease.

[0077] S4: Integrate the denoised image, color features, texture features, vegetation index features and segmentation masks, and build and train a pine wood nematode segmentation model based on U-Net;

[0078] S41: Integrate the denoised image, color features, texture features, vegetation index features and the segmentation mask in step S3 into a training set. Each sample in the training set is specifically:

[0079] ;

[0080] in, is the input image, including the denoised image , Color Features , Contrast ,energy and Normalized Difference Vegetation Index ;

[0081] S42: Building a pine wilt disease segmentation model based on the U-Net architecture. The pine wilt disease segmentation model consists of an encoder and a decoder, and a jump connection is added between the encoder and the decoder.

[0082] S43: Use the training set data to train the pine wood nematode segmentation model and use the stochastic gradient descent algorithm to minimize the loss function , a trained pine wilt disease segmentation model is obtained, and the loss function is specifically:

[0083] ;

[0084] in, is the segmentation mask predicted by the pine wilt segmentation model At pixel position The value at ; and They are the denoised images respectively. The number of pixels in the horizontal direction and the number of pixels in the vertical direction.

[0085] This step uses the U-Net architecture to build and train a segmentation model by integrating multiple feature images with segmentation masks marked by professionals. This method makes full use of the color, texture, and vegetation index features of multispectral images, combined with high-precision manually annotated data, to enhance the model's learning ability and segmentation accuracy. By minimizing the loss function, the model can more effectively identify areas infected by pine wood nematode disease, thereby improving the reliability and robustness of overall recognition.

[0086] S5: Use the multispectral camera carried by the drone to collect new tree area images, denoise the new tree area images and extract color features, texture features and vegetation index features; apply the trained pine wilt disease segmentation model to identify the pine wilt disease infected area:

[0087] Using a multispectral camera carried by a drone to collect a new tree area image, performing denoising on the new tree area image in step S1 to obtain a denoised image, and then performing color feature, texture feature and vegetation index feature extraction on the denoised image in step S2; integrating the denoised image obtained using the new tree area image with the color feature, texture feature and vegetation index feature to obtain a new input image;

[0088] The trained pine wilt disease segmentation model is used to segment the new input image. The pine wilt disease segmentation model outputs a predicted segmentation mask. The pixel position with a pixel value of 1 in the segmentation mask is the identified pine wilt disease infected area.

[0089] The pixels with a pixel value of 1 in the segmentation mask are set to white, and the pixels with a pixel value other than 1 are set to black. The result is as follows Figure 2 shown.

[0090] Embodiment 2: The present invention also discloses an intelligent identification system for pine wood nematode disease based on drone technology, comprising the following five modules:

[0091] Collection and denoising module: Use a drone equipped with a multispectral camera to collect images of the tree area, and apply a bilateral filtering algorithm to denoise the images;

[0092] Feature extraction module: extracts color features, texture features and vegetation index features from the denoised image;

[0093] Annotation module: manually annotate the denoised image and mark the areas infected with pine wood nematode disease;

[0094] Model building module: Integrate denoised images, color features, texture features, vegetation index features and segmentation masks, build and train a pine wood nematode segmentation model based on U-Net;

[0095] Model application module: Use the trained pine wilt disease segmentation model to identify pine wilt disease infected areas.

[0096] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0097] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0098] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An intelligent identification method for pine wood nematode disease based on drone technology, characterized in that: The following steps are involved: S1: Use a multispectral camera mounted on a drone to collect images of the tree area, apply a bilateral filtering algorithm to denoise the image, and obtain a denoised image; S2: Extract color features, texture features and vegetation index features from the denoised image; the texture features include contrast and energy ; The vegetation index feature is the normalized difference vegetation index ; S3: Manually annotate the denoised image, mark the areas infected with pine wilt disease, and generate segmentation masks ,in , Indicates pixel position It is an area infected with pine wood nematode disease. Indicates pixel position It is not an area infected by pine wood nematode disease; S4: Integrate the denoised image, color features, texture features, vegetation index features and segmentation masks, and build and train a pine wood nematode segmentation model based on U-Net; The step S4 includes the following steps: S41: Integrate the denoised image, color features, texture features, vegetation index features and the segmentation mask in step S3 into a training set. Each sample in the training set is specifically: ； in, is the input image, including the denoised image , Color Features , Contrast ,energy and Normalized Difference Vegetation Index ; S42: Building a pine wilt disease segmentation model based on the U-Net architecture. The pine wilt disease segmentation model consists of an encoder and a decoder, and a jump connection is added between the encoder and the decoder. S43: Use the training set data to train the pine wood nematode segmentation model and use the stochastic gradient descent algorithm to minimize the loss function , obtain the trained pine wood nematode disease segmentation model; S5: Use the multispectral camera carried by the drone to collect new tree area images, denoise the new tree area images and extract color features, texture features and vegetation index features; apply the trained pine wilt disease segmentation model to identify the pine wilt disease infected area.

2. The intelligent identification method of pine wood nematode disease based on drone technology according to claim 1 is characterized in that: The step S1 includes the following steps: S11: Using a multispectral camera mounted on a drone to capture images of tree areas ; S12: Apply bilateral filtering algorithm to the image of tree area Perform denoising to obtain the denoised image , specifically: ； in, and Images of tree areas The horizontal pixel position and the vertical pixel position of The denoised image At pixel position The pixel value at ; Image of the tree area At pixel position The pixel value at ; and are the horizontal and vertical offsets respectively, and Traverse within the filter window; is the size of the filter window; Image of the tree area At pixel position The pixel value at ; is the spatial Gaussian kernel function; is the range Gaussian kernel function.

3. The intelligent identification method of pine wood nematode disease based on drone technology according to claim 2 is characterized in that: The step S2 includes the following steps: S21: Use color space transformation to transform the denoised image The red, green, and blue bands in the image are transformed from the RGB color space to the HSV color space to obtain color features. ; S22: extract texture features; S23: Extracting vegetation index features, where the vegetation index features are normalized difference vegetation index , specifically: ； in, Normalized Difference Vegetation Index At pixel position The value at and They are the denoised images respectively. Near infrared and red light bands; and They are and At pixel position The pixel value at .

4. The intelligent identification method of pine wood nematode disease based on drone technology according to claim 3 is characterized in that: The step S22 includes the following steps: S221: For each pixel position , construct the local gray-level co-occurrence matrix , specifically: ； in, is the element index in the local gray-level co-occurrence matrix, and Corresponding to the two gray levels in the denoised image, ranging from 0 to An integer between is the number of gray levels; At pixel position of In the neighborhood, the gray level is and The frequency of pixel pairs; For Centered A collection of pixel position indices within the range. for The pixel position index in ; The denoised image At pixel position The pixel value at ; is the indicator function, when 1 when it is, otherwise 0; S222: Extracting texture features based on the local gray level co-occurrence matrix, wherein the texture features include contrast and energy , specifically: ； ； in, For contrast At pixel position The contrast value at For energy At pixel position The energy value at .

5. The intelligent identification method of pine wood nematode disease based on drone technology according to claim 4 is characterized in that: In step S4, the loss function is specifically: ； in, is the segmentation mask predicted by the pine wilt segmentation model At pixel position The value at ; and They are the denoised images respectively. The number of pixels in the horizontal direction and the number of pixels in the vertical direction.

6. The intelligent identification method of pine wood nematode disease based on drone technology according to claim 5 is characterized in that: The step S5 includes the following steps: Using a multispectral camera carried by a drone to collect a new tree area image, performing denoising on the new tree area image in step S1 to obtain a denoised image, and then performing color feature, texture feature and vegetation index feature extraction on the denoised image in step S2; integrating the denoised image obtained using the new tree area image with the color feature, texture feature and vegetation index feature to obtain a new input image; The trained pine wilt disease segmentation model is used to segment the new input image. The pine wilt disease segmentation model outputs a predicted segmentation mask. The pixel position with a pixel value of 1 in the segmentation mask is the identified pine wilt disease infected area.

7. An intelligent identification system for pine wood nematode disease based on drone technology, characterized in that: include: Collection and denoising module: Use a drone equipped with a multispectral camera to collect images of the tree area, and apply a bilateral filtering algorithm to denoise the images; Feature extraction module: extracts color features, texture features and vegetation index features from the denoised image; Annotation module: manually annotate the denoised image and mark the areas infected with pine wood nematode disease; Model building module: Integrate denoised images, color features, texture features, vegetation index features and segmentation masks, build and train a pine wood nematode segmentation model based on U-Net; Model application module: Use the trained pine wilt disease segmentation model to identify pine wilt disease infected areas; To realize the intelligent identification method of pine wood nematode disease based on drone technology as described in any one of claims 1-6.

Images